WO2016133121A1 - Abnormality diagnosis method and abnormality diagnosis system - Google Patents

Abstract

The present invention is provided with: a model preparation step for preparing a simulation model (3) of an object to be monitored (2); an operation start step for starting operation of the monitoring object; a measurement step for measuring the internal state quantity during the state in which the object to be monitored (2) is in operation and extracting an actual measurement value (xˆ); a prediction step for inputting into the simulation model (3) a control input value (u) that is the same as the state in which the object to be monitored (2) is in operation, and calculating a prediction value (x) of the internal state quantity of the object to be monitored (2); a Mahalanobis-distance calculation step for calculating the Mahalanobis distance (MD) based on the difference between the actual measurement value (xˆ) and the prediction value (x); and an abnormality diagnosis step for diagnosing whether the state in which the article to be monitored (2) is in operation is abnormal on the basis of the Mahalanobis distance (MD).

Description

Abnormality diagnosis method and abnormality diagnosis system

The present disclosure relates to an abnormality diagnosis method and an abnormality diagnosis system, and more particularly, to an abnormality diagnosis method and an abnormality diagnosis system suitable for finding an abnormality in an unsteady state having a dynamic change.

In the field of various turbines such as gas turbine power plants, nuclear power plants, thermal power plants, and internal combustion engines such as jet engines, operating conditions (including tests) are monitored and abnormal to maintain stable operation and output. Diagnosis is done.

For example, Patent Document 1 discloses a monitoring unit that acquires predetermined monitoring target data from a monitoring target, calculates the Mahalanobis distance to detect an abnormality of the monitoring target, and a monitoring target related to an abnormal signal indicating a sign of abnormality. Monitoring a predictor of an abnormality in the monitoring target, comprising: a data processing unit that extracts a related signal that is data and generates a predetermined input signal; and a failure diagnosis unit that performs a fault diagnosis of the monitoring target based on the input signal A monitoring system that can automate a series of processes from fault diagnosis to failure diagnosis is disclosed.

Patent Document 2 discloses an accumulated data storage unit for storing accumulated data including values of variables input in the past, a newly input value for each variable, and a maximum value and a minimum value for a predetermined period included in the accumulated data. A determination unit that extracts a value and determines an intermediate value as a median value; a first calculation unit that calculates a difference value between a newly input value and a median value for each variable; A second calculation unit that obtains the Mahalanobis distance using data of a predetermined unit space, and a determination unit that determines whether the Mahalanobis distance is within a predetermined threshold range and diagnoses an abnormality An abnormality diagnosing device for diagnosing an abnormality of the plant by comparing values of a plurality of variables input to a predetermined unit space is disclosed.

JP 2011-090382 A JP 2014-035282 A

By the way, a monitoring object such as a plant or an internal combustion engine generally has a steady state indicating a stable operation state and an unsteady state indicating a transient unstable operation state up to the steady state. Yes. The unsteady state is different depending on the environmental condition and the operating condition at that time even for the same monitoring object, and hardly shows the same dynamic change.

In the monitoring system described in Patent Document 1, a failure diagnosis input signal is generated from an abnormality signal indicating a sign of abnormality and a related signal by calculating a Mahalanobis distance of monitoring target data. Here, after calculating the Mahalanobis distance of the monitoring target data, it is necessary to prepare reference data in advance in order to determine whether an abnormality or a sign of the abnormality is indicated. At this time, in the steady state, since the operation state and the output state are stable, it is possible to prepare the reference data. However, in the non-steady state with dynamic change, the reference data cannot be created only from the monitoring target data, and abnormality diagnosis cannot be performed.

Also, in the abnormality diagnosis apparatus described in Patent Document 2, since the Mahalanobis distance is calculated using past accumulated data, as in Patent Document 1, it is compared with past data in a steady state. Although the abnormality can be diagnosed, the abnormality cannot be diagnosed in the non-steady state.

The present disclosure has been devised in view of the above-described problems, and provides an abnormality diagnosis method and an abnormality diagnosis system capable of diagnosing an abnormality not only in a steady state but also in an unsteady state of an object to be monitored. With the goal.

A first aspect of the present disclosure is a method for diagnosing an abnormality of a monitoring object including a non-steady-state operation state, the model creating step of creating a simulation model of the monitoring object, and the operation state of the monitoring object A measurement step of measuring an internal state quantity in the system and extracting an actual measurement value, and inputting a control input value identical to the operation state of the monitoring target object to the simulation model to obtain a predicted value of the internal state quantity of the monitoring target object A prediction step to calculate, a Mahalanobis distance calculation step to calculate a Mahalanobis distance from the difference between the actual measurement value and the prediction value, and whether or not the operating state of the monitoring object is abnormal based on the Mahalanobis distance An abnormality diagnosis step is provided.

The Mahalanobis distance calculating step may include a step of calculating an error vector having the difference and the integral value as components. Furthermore, the prediction step may calculate the prediction value based on a previous actual measurement value in time series.

In addition, a second aspect of the present disclosure is an abnormality diagnosis system for a monitoring object including an unsteady operation state, and includes a simulation model that simulates the monitoring object and an internal state of the monitoring object in the operation state. A measuring means for measuring a state quantity; and a Mahalanobis distance is calculated from a difference between a predicted value obtained by the simulation model and an actual value extracted from the measuring means, and the monitoring object is operated based on the Mahalanobis distance. The gist is provided with a diagnostic device for diagnosing whether or not the state is abnormal and a control device that transmits at least the same monitoring input value to the monitoring object and the simulation model.

The diagnostic apparatus may calculate the Mahalanobis distance based on an error vector having the difference and the integral value as components. Further, the simulation model may calculate the predicted value based on a previous measured value in time series. The monitoring object is, for example, a reusable spacecraft engine.

According to the abnormality diagnosis method and abnormality diagnosis system according to the present disclosure, a simulation model that simulates the internal state of the monitoring target is created, and the monitoring target is calculated using the difference between the actual measurement value of the monitoring target and the predicted value of the simulation model. Since the presence or absence of abnormality of the object is diagnosed, the simulation model can calculate the predicted value adapted to the environmental conditions and operating conditions at the time of abnormality diagnosis, and the measured value of the monitored object by taking the difference Can be replaced with a variation value with respect to a normal value. Therefore, even when the monitoring object is in an unsteady state, it is possible to respond to the dynamic change and diagnose abnormality not only in the steady state but also in the unsteady state of the monitoring object. be able to. Further, by using the Mahalanobis distance for abnormality diagnosis, it is possible to simplify and speed up the abnormality diagnosis.

FIG. 1 is a schematic overall configuration diagram illustrating an abnormality diagnosis system according to the present disclosure. FIG. 2 is a flowchart showing the abnormality diagnosis method according to the present disclosure. FIGS. 3A and 3B are explanatory diagrams of the Mahalanobis distance calculation step, FIG. 3A shows an error vector, and FIG. 3B shows an example of a prediction value calculation method. . 4 (a) and 4 (b) are explanatory diagrams of the abnormality diagnosis step, FIG. 4 (a) is a conceptual diagram of Mahalanobis distance, and FIG. 4 (b) is a conceptual diagram of abnormality diagnosis. . 5 (a) to 5 (c) are explanatory diagrams for verifying the effectiveness when the present disclosure is applied to a reusable spacecraft engine. FIG. 5 (a) shows control input values, FIG. 5B shows simulated data of actual measurement values, and FIG. 5C shows an abnormality diagnosis result based on the Mahalanobis distance.

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. Here, FIG. 1 is a schematic overall configuration diagram illustrating the abnormality diagnosis system according to the present disclosure. FIG. 2 is a flowchart showing the abnormality diagnosis method according to the present disclosure. FIGS. 3A and 3B are explanatory diagrams of the Mahalanobis distance calculation step, FIG. 3A shows an error vector, and FIG. 3B shows an example of a prediction value calculation method. . 4 (a) and 4 (b) are explanatory diagrams of the abnormality diagnosis step, FIG. 4 (a) is a conceptual diagram of Mahalanobis distance, and FIG. 4 (b) is a conceptual diagram of abnormality diagnosis. .

As shown in FIG. 1, the abnormality diagnosis system 1 according to an embodiment of the present disclosure is an abnormality diagnosis system for a monitoring object 2 including an unsteady operation state, and simulates the monitoring object 2. The model 3, the measuring means 4 for measuring a predetermined internal state quantity in the operating state of the monitored object 2, the predicted value x obtained by the simulation model 3, and the actual measurement value x ^ extracted from the measuring means 4 (actual The notation is calculated on the x from the difference (x ^ -x) from ^ (circumflex, hat) on the x (the same applies hereinafter), and the operating state of the monitored object 2 is based on the Mahalanobis distance MD. A diagnosis device 5 for diagnosing whether there is an abnormality, and a control device 6 for transmitting the same control input value u to the monitored object 2 and the simulation model 3 are provided.

The monitoring object 2 is, for example, an engine for a reusable spacecraft, but is not limited to this. Other internal combustion engines such as a jet engine, gas turbine power plant, nuclear power plant, thermal power plant It may be various plants such as a chemical plant. In particular, the monitoring object 2 preferably includes a steady state indicating a stable operating state and an unsteady state indicating a transient and unstable operating state up to the steady state in the operating state.

The simulation model 3 is a model that can estimate the internal state quantity of the monitored object 2, and is created by applying a numerical simulation technique, for example. When creating a simulation model, it may be described in a recursive expression (ARMA) in consideration of real-time processing. For example, when the monitoring object 2 is a reusable spacecraft engine, the internal state quantities include, for example, the combustion pressure Pc, the regeneration cooling outlet temperature Tjmf, the fuel pump rotational speed Nf, the oxidant pump rotational speed No, the fuel Pump outlet pressure Pdf, oxidant pump outlet pressure Pdo, etc. are selected. Therefore, a simulation model capable of calculating these internal state quantities is created. The simulation model 3 may be a single simulation model that simulates the entire monitoring target 2 or may be constructed by a plurality of simulation models that can individually calculate the internal state quantities.

The measuring means 4 is installed on the monitoring object 2, and for example, combustion pressure Pc, regeneration cooling outlet temperature Tjmf, fuel pump rotational speed Nf, oxidant pump rotational speed No, fuel pump outlet pressure Pdf, oxidant pump outlet pressure Pdo. It is a sensor which measures internal state quantities, such as. For example, the measuring unit 4 is a pressure gauge, a thermometer, a rotary encoder, or the like, but is not limited thereto, and is appropriately selected depending on the amount of internal state to be measured by the monitoring object 2.

The control device 6 is a device that transmits a control input value u necessary for operating the monitoring object 2 to the monitoring object 2. The operation status of the monitoring object 2 may be actual operation or experimental. In addition, the control device 6 also transmits a control input value u necessary for the operation of the monitored object 2 to the simulation model 3. The simulation model 3 calculates an internal state quantity based on the control input value u, and calculates a predicted value x for each internal state quantity. Note that the output value y of the monitored object 2 driven by the control input value u may be measured and extracted to the outside.

The diagnostic device 5 receives the data of the actual measurement value x ^ measured by the measuring means 4 and the data of the predicted value x calculated by the simulation model 3, and uses these data to detect abnormalities in the monitored object 2 It is a device that performs diagnosis. In the diagnostic device 5, for example, the processing is performed based on the flowchart illustrated in FIG. 2. The diagnosis result and diagnosis data may be externally output from the diagnosis device 5 by monitor output, paper output, data output, or the like.

Here, the flowchart shown in FIG. 2 includes a model creation step (Step 1) for creating the simulation model 3 of the monitoring object 2, an operation start step (Step 2) for starting the operation of the monitoring object 2, and the monitoring object. The measurement step (Step 3) for measuring the internal state quantity in the operating state of the object 2 and extracting the actual measurement value x ^, and the control input value u identical to the operating state of the monitoring object 2 is input to the simulation model 3 and monitored A prediction step (Step 4) for calculating the predicted value x of the internal state quantity of the object 2 and a Mahalanobis distance calculation step for calculating the Mahalanobis distance MD from the difference (x ^ -x) between the actual measurement value x ^ and the predicted value x ( Step 5) and an abnormality diagnosis step for diagnosing whether or not the operation state of the monitoring object 2 is abnormal based on the Mahalanobis distance MD (Step 6) It has a, and.

In the diagnosis device 5 described above, the Mahalanobis distance calculation step (Step 5) and the abnormality diagnosis step (Step 6) are performed. In the abnormality diagnosis method according to the present embodiment, whether or not the obtained data (actual measurement value x ^) is abnormal is diagnosed based on multivariate analysis using the Mahalanobis distance. By using this Mahalanobis distance, the correlation of multiple variables can be processed at once, eliminating the need to diagnose whether each variable is abnormal individually, simplifying and speeding up abnormality diagnosis Can be achieved.

Further, as shown in FIG. 2, the Mahalanobis distance calculating step (Step 5) includes a difference calculating step (Step 51) for calculating a difference (x ^ −x) between the actual measurement value x ^ and the predicted value x, and a difference (x An error vector calculation step (Step 52) for calculating an error vector ε having components of ^ −x) and an error integral value Σε, and a Mahalanobis distance calculation step (Step 53) for calculating the Mahalanobis distance MD based on the error vector ε. , May be included.

Here, the error vector ε can be expressed, for example, as shown in FIG. The integral value Σε constituting one component of the error vector ε can be calculated as a so-called integral value by continuously calculating an error vector that changes from moment to moment, and the error vector ε every fixed time (span). Is calculated as the sum of the differences (x ^ -x). Thus, by using the error (difference) integral value Σε, it is possible to prevent the accumulated error evaluation sensitivity in the same direction from becoming weak.

For example, when the combustion pressure Pc, the regeneration cooling outlet temperature Tjmf, the fuel pump rotational speed Nf, the oxidant pump rotational speed No, the fuel pump outlet pressure Pdf, and the oxidant pump outlet pressure Pdo are selected as the internal state quantities, an error occurs. As shown in FIG. 3A, the vector ε can be expressed as a matrix of (ΔPc, ΔTjmf, ΔNf, ΔNo, ΔPdf, ΔPdo, ΣΔPc, ΣΔTjmf, ΣΔNf, ΣΔNo, ΣΔPdf, ΣΔPdo). In this case, since the error vector ε includes 12 variables, the error vector ε is included in the vector space R ¹² formed by these variables.

The prediction step (Step 4) includes an input step (Step 41) for inputting the same control input value u to the operation of the monitored object 2 to the simulation model 3, and a predicted value x of the internal state quantity based on the control input value u. A predicted value calculating step (Step 42). In the predicted value calculation step Step 42 (prediction step (Step 4)), as shown in FIG. 3B, the predicted value xn is calculated based on the previous measured value x _n−1 ^ in time series. It may be. That is, the predicted value xn is calculated based on the actual measurement value x _n-1 ^, and the predicted value x _{n + 1} is calculated based on the actual measurement value x _n ^. Such processing can suppress accumulation of errors, improve the accuracy of the predicted value _xn , and improve the accuracy of abnormality diagnosis.

In order to calculate the Mahalanobis distance MD from the error vector ε in the Mahalanobis distance calculation step (Step 53), first, the error vector ε is normalized using Equation 1 and converted into a state independent of the physical quantity unit. For normalization of the error vector ε, the total average vector during the operation period

And deviation

Is used.

... (Formula 1)
However,

It is. Then, the error vector εn ″ standardized based on Equation 1 is again expressed as εn and used for the subsequent calculation.

Next, the Mahalanobis distance MD is calculated using Formula 2. Here, ε ^T means a transposed matrix of the error vector ε, and dim (ε) means the dimension of the error vector ε. In addition, the covariance matrix can be derived from past accumulated data diagnosed as normal, for example.

... (Formula 2)
However,

Is the covariance matrix.

By calculating the Mahalanobis distance MD and connecting the equidistant points, for example, the correlation of the internal state quantities as shown in FIG. 4A can be obtained, and as the distance from the center of the illustrated elliptical region increases The error is large, and it can be diagnosed as abnormal when deviating from this region. Here, the correlation shown in FIG. 4A shows the correlation of only two variables of the internal state quantities D1 and D2 in order to promote intuitive understanding. According to this correlation, it can be understood that the allowable amount of error is large in the major axis direction of the substantially elliptical region, and the allowable amount of error is small in the minor axis direction of the substantially elliptical region. Although not shown, as described above, when 12 variables are used, a 12-dimensional correlation is obtained.

In the abnormality diagnosis step Step 6, for example, as shown in FIG. 4B, for the error (difference value) that changes every moment, the Mahalanobis distance MD is calculated for each diagnosis, and each time the error (difference) is calculated. Value) is within the range of Mahalanobis distance MD. For example, the Mahalanobis distance MD1 at time t1, the Mahalanobis distance MD2 at time t2, the Mahalanobis distance MD3 at time t3, the Mahalanobis distance MD4 at time t4, and the Mahalanobis distance MD5 at time t5 are changed from time to time depending on the environmental conditions and operating conditions at that time. It will change. Note that the diagram illustrated in FIG. 4B is illustrated to facilitate intuitive understanding of the abnormality diagnosis method according to the present embodiment.

According to the abnormality diagnosis method and abnormality diagnosis system 1 according to the present embodiment described above, the simulation model 3 that simulates the internal state of the monitored object 2 is created, and the actual measurement value x ^ of the monitored object 2 and the simulation model 3 Since the presence or absence of abnormality of the monitored object 2 is diagnosed using the difference (x ^ -x) from the predicted value x, the predicted value adapted to the environmental conditions and operating conditions at the time of abnormality diagnosis by the simulation model 3 x can be calculated, and by taking the difference, the actual measurement value x ^ of the monitoring object 2 can be replaced with a fluctuation value with respect to the normal value. Therefore, even when the operation state of the monitored object 2 is an unsteady state, it is possible to respond to the dynamic change, and not only the steady state but also the unsteady state of the monitored object 2 is abnormal. Can be diagnosed.

Here, FIG. 5A to FIG. 5C are explanatory diagrams for verifying the effectiveness when the present disclosure is applied to an engine for a reusable spacecraft, and FIG. FIG. 5B shows simulated data of actually measured values, and FIG. 5C shows an abnormality diagnosis result based on Mahalanobis distance. In FIGS. 5A and 5B, the numerical value of thrust is indicated by a solid line, the numerical value of fuel is indicated by a dotted line, the numerical value of oxidant is indicated by a one-dot chain line, and the numerical value of combustion pressure is indicated by a two-dot chain line. In FIG. 5A, the portion where the thrust is convex upward (substantially trapezoidal portion) simulates an unsteady state.

In order to obtain the thrust shown in FIG. 5 (a), the amount of fuel and oxidant shall be controlled as shown. Now, in order to verify the effectiveness of the abnormality diagnosis based on the Mahalanobis distance MD, as shown in FIG. 5B, an offset value (α portion in the figure) is given to the normal measured value, The simulation data of the actual measurement value that intentionally included the numerical value was created. And when the process of the Mahalanobis distance calculation step Step5 mentioned above was performed using the simulation value of this measured value and the predicted value calculated | required by the simulation model 3, the result shown in FIG.5 (c) was obtained.

In FIG. 5 (c), the solid line indicates the temporal change of the Mahalanobis distance MD, and the black circle in the figure indicates the point diagnosed as abnormal. According to this verification result, it can be understood that the Mahalanobis distance MD of the part corresponding to the offset part to which an intentionally abnormal numerical value is given is diagnosed as abnormal. Therefore, it is recognized that the abnormality diagnosis method and abnormality diagnosis system 1 according to the present embodiment have responsiveness capable of performing abnormality diagnosis with respect to an operation state including an unsteady state.

The present disclosure is not limited to the above-described embodiment, and various changes can be made without departing from the spirit of the present disclosure.

Claims (7)

1. An abnormality diagnosis method for an object to be monitored including an unsteady operation state,
   A model creation step of creating a simulation model of the monitored object;
   A measurement step of measuring an internal state quantity in an operation state of the monitoring object and extracting an actual measurement value;
   A prediction step of inputting the same control input value as the operation state of the monitoring object to the simulation model and calculating a predicted value of the internal state quantity of the monitoring object;
   A Mahalanobis distance calculating step for calculating a Mahalanobis distance from a difference between the actual measurement value and the predicted value;
   An abnormality diagnosis step of diagnosing whether or not the operating state of the monitoring object is abnormal based on the Mahalanobis distance;
   An abnormality diagnosis method comprising:
2. The abnormality diagnosis method according to claim 1, wherein the Mahalanobis distance calculating step includes a step of calculating an error vector having the difference and an integral value thereof as components.
3. 3. The abnormality diagnosis method according to claim 2, wherein the predicting step calculates the predicted value based on a previous measured value in time series.
4. An abnormality diagnosis system for an object to be monitored including an unsteady operation state,
   A simulation model simulating the monitored object;
   Measuring means for measuring an internal state quantity in an operating state of the monitoring object;
   A Mahalanobis distance is calculated from a difference between a predicted value obtained by the simulation model and an actual measurement value extracted from the measuring means, and whether or not the operating state of the monitored object is abnormal is calculated based on the Mahalanobis distance. A diagnostic device to
   A control device that transmits the same control input value to at least the monitoring object and the simulation model;
   An abnormality diagnosis system comprising:
5. The abnormality diagnosis system according to claim 4, wherein the diagnosis device calculates the Mahalanobis distance based on an error vector having the difference and an integral value as a component.
6. The abnormality diagnosis system according to claim 5, wherein the simulation model calculates the predicted value based on a previous measured value in time series.
7. The abnormality diagnosis system according to claim 4, wherein the monitoring object is an engine for a reusable spacecraft.

Images